Can't get inpainting to work on v4.1, Searge. I did as instructed, loaded the image in the source image and masking boxes. Did a mask, and it outputs the same image. How can I get it to work?

Hi Searge! Awesome workflow, I wanted to ask how to get the best results with controlnet turned on (I am using the intermediate script)? The outputs are great with it off, however once its turned on, the higher the strength the less realistic the image (but the more it resembles the canny of the input). Do you have recommended settings for using controlnet?

Ran the installer script, but having the following error: Error occurred when executing SeargeMagicBox: get_additional_models() takes 2 positional arguments but 3 were given

Nice workflow

When I add another type of custom node to Comfyai for example - reactor node I can no longer use this one as every node goes red saying it's not installed. I have to delete the reactor folder from the custom node folder for it to work again. Is there any way around this?

how should i use this with a "stability matrix" ComfyUI install? not sure if portable or whatever.

Am I the only one who's VRAM gets maxed out with the latest v4 workflows? On v3 of the workflow I could easily do 1024x1024 with 2x hires in around 65 seconds using around 14-16 gigs of VRAM.

With the latest workflows, it takes me at minimum 120 seconds to do a 1024x1024 with 1.25x hires, and doing so results in my 24 gigs of VRAM being maxed out and overflowing into shared memory.

The images are undoubtedly higher quality, but I don't know if it's higher enough to justify waiting twice as long.

The workflow is really fast and amazing with SDXL checkpoint. But whenever I attempt to run an SD checkpoint, I receive an error message, whether I have made a mistake or i am simply dumb (I am dumb). Tried to reinstall the XDSL UI but still the same. I don't have the portable comfy, so I copied with git and downloaded/copied the recommend and required file in the right folder, run the python -m pip install opencv-python command but always the same error.

https://i.ibb.co/rdS4c15/Capture-d-cran-2023-09-11-121750.png

Are inpainting and img2img working properly? No matter what combination of settings I try on 4.1 starting from the default settings in the workflow JSON, I just get a version of the original sketch back out where the inpainted area appears to be inferred from the image, not conditioned on the prompt.

To reproduce:
• Pull Searge and Comfy repos to make sure they're up to date
• Open Searge-SDXL-Reborn-v4_1-hq.json
• Change Operating Mode to 'in-painting' or 'image-to-image'
• Open Inpainting Mask in Mask Editor and remove some portion of the image
• Confirm all model settings point to corresponding files on disk
• Change image size preset to 1:1 (1024x1024)
• Queue Prompt

Expected result:
I'd expect the inpainted area to be an image of the scene described in the prompt(s)

Result:

The inpainted area is a continuation of the source image and ignores the prompts. For example - mask out the girl's face in the default sketch and it will draw her face back in with some deviation from the original.

Thanks, great workflow! I'm still overwhelmed with all the possibilities, nevertheless looking forward to new updates and implementations.

Hi,

would it be possible to get an optional workflow with the PerpNeg sampler integrated? The tradeoff here is nearly double generation times for better prompt adherence

https://github.com/bvhari/ComfyUI_PerpNeg

Hello sir, Really loved your workflow. I had like to request. I don't have a huge vram so I usually upscale my photos using Ultimate SD tile upscale and I use SD 1.5 models. If I was to incorporate that kind of workflow into your workflow for upscaling. How can I achieve it? And I am really sorry if this is not right of me to ask to make modifications to your workflow.

How exactly does the latent detailer process work? I have zero programming experience, so I have only deduced that it upscales the sampled latent by 2x, runs 5 steps, rescales to original, and runs another 10 steps.

I can't decipher the part of the code that sets the step counts:
start_step=int(100 - percent * 2), last_step=int(100 - percent)

I'm trying to deseargify your workflow to better understand SD, your WF, and relevant systems.

Hi Searge. Where or how should I find or enter this code? --disable-smart-memory

Hi Searge,

Love the work you have done.... I can achieve wayyy better images using this workflow. the fact i can create my own "workspaces" by moving everything around to suit myself to what ever im doing is great.... so great. I save my basic img gen workspace in a png and then if i want controlnet i get my controlnet png and drag that in.... amazing work!!!! Comfy was a little overwhelming coming from A1111 and i was at a loss as to how to setup all the different things to work correctly (controlnet for example).

The only thing i miss is a "testing station". Adding a X/Y plot area would be amazing (somehow having a X/Y/Z would be even better ;) )

Keep up the good work man. This must be quite the project for you, especially with everything always changing. Most of us are benefiting for free. I dont know your situation but please dont worry about any pressure from the community (like above), try not to burnout doing this work.

Sorry to inform you that mine crashes in sampling stage.

custom_sdxl_ksampler.py, line 202

if refiner_detail_boost > 0.0

TypeError: '>' not supported between instances of 'NoneType' and 'float'

I've all models in place. Thanks a lot.

Can anybody help? the past version worked fine (3.x), but the new 4.1 is not working and i'm getting an error, I updated searge with the new script. The error says:

"Error occurred when executing SeargeMagicBox: Error while deserializing header: MetadataIncompleteBuffer File ":\xxxxx\ComfyUI_windows_portable_nvidia_cu118_or_cpu\ComfyUI_windows_portable\ComfyUI\execution.py", line 152, in recursive_execute output_data, output_ui = get_output_data(obj, input_data_all) File "F:\xxxxx\ComfyUI_windows_portable_nvidia_cu118_or_cpu\ComfyUI_windows_portable\ComfyUI\execution.py", line 82, in get_output_data return_values = map_node_over_list(obj, input_data_all, obj.FUNCTION, allow_interrupt=True) File ":\xxxxx\ComfyUI_windows_portable_nvidia_cu118_or_cpu\ComfyUI_windows_portable\ComfyUI\execution.py", line 75, in map_node_over_list results.append(getattr(obj, func)(**slice_dict(input_data_all, i))) File "F:\xxxxx\ComfyUI_windows_portable_nvidia_cu118_or_cpu\ComfyUI_windows_portable\ComfyUI\custom_nodes\SeargeSDXL\modules\magic_box.py", line 269, in process (data, stage_result) = self.run_stage(stage, data, stage_input) File ":\xxxxx\ComfyUI_windows_portable_nvidia_cu118_or_cpu\ComfyUI_windows_portable\ComfyUI\custom_nodes\SeargeSDXL\modules\magic_box.py", line 248, in run_stage (data, stage_result) = stage_processor.process(data, stage_input) File ":\xxxxx\ComfyUI_windows_portable_nvidia_cu118_or_cpu\ComfyUI_windows_portable\ComfyUI\custom_nodes\SeargeSDXL\modules\stage_load_checkpoints.py", line 211, in process clip_vision_model = NodeWrapper.clipvision_loader.load_clip(clipvision_name)[0] File ":\xxxxx\ComfyUI_windows_portable_nvidia_cu118_or_cpu\ComfyUI_windows_portable\ComfyUI\nodes.py", line 757, in load_clip clip_vision = comfy.clip_vision.load(clip_path) File ":\xxxxx\ComfyUI_windows_portable_nvidia_cu118_or_cpu\ComfyUI_windows_portable\ComfyUI\comfy\clip_vision.py", line 110, in load sd = load_torch_file(ckpt_path) File ":\xxxxx\ComfyUI_windows_portable_nvidia_cu118_or_cpu\ComfyUI_windows_portable\ComfyUI\comfy\utils.py", line 13, in load_torch_file sd = safetensors.torch.load_file(ckpt, device=device.type) File ":\xxxxx\ComfyUI_windows_portable_nvidia_cu118_or_cpu\ComfyUI_windows_portable\python_embeded\lib\site-packages\safetensors\torch.py", line 309, in load_file with safe_open(filename, framework="pt", device=device) as f:"

This is a great workflow. Thank you for creating it. One question: is there a way to do live previews of the steps of the image generation in ComfyUI (meaning to see what it looks like in step 12, 15, etc.)? I suspect that in some cases the earlier steps look better- and it'd help in assessing whether to limit the steps in which the controlnets are applied. Right now it is a bit of blind guessing.

first: while 4.0 works great without any issues at all, 4.1 always stuck's at highres detailer stage, takes ages and then -literally- crushes my mashine ;)

second: sdxl has two different types to prompt ... i think clip l and clip g or something was it,

but ... what exactly is "main prompt" and "secondary prompt" in this workflow refering to?

I installed it twice from zero but I still get this error;

Help please

Im a bit late joining the ComfyUI's side but wanted to ask something. Is there any workflow that allows me:
- Generate a SDXL output AND automatically use this output via Img2img in a custom SD 1.5 I want, with a custom upscaler and custom VAE? All I see is generate SDXL output and change mode to img2img and load the image but I'd want to do that automatically for running batches with a custom SD 1.5 model in the img2img and a sharper VAE for that. IS this possible?

I am running on Linux, the installation has failed many times, the node has not been loaded out, please ask the author has any solution, thank you very much

when I try to do an inpaint, the error below occurs, regardless of the workflow settings:

Error occurred when executing SeargeMagicBox: too many values to unpack (expected 2)

File "C:\ComfyUI_windows_portable\ComfyUI\execution.py", line 152, in recursive_execute output_data, output_ui = get_output_data(obj, input_data_all) File "C:\ComfyUI_windows_portable\ComfyUI\execution.py", line 82, in get_output_data return_values = map_node_over_list(obj, input_data_all, obj.FUNCTION, allow_interrupt=True) File "C:\ComfyUI_windows_portable\ComfyUI\execution.py", line 75, in map_node_over_list results.append(getattr(obj, func)(**slice_dict(input_data_all, i))) File "C:\ComfyUI_windows_portable\ComfyUI\custom_nodes\SeargeSDXL\modules\magic_box.py", line 269, in process (data, stage_result) = self.run_stage(stage, data, stage_input) File "C:\ComfyUI_windows_portable\ComfyUI\custom_nodes\SeargeSDXL\modules\magic_box.py", line 248, in run_stage (data, stage_result) = stage_processor.process(data, stage_input) File "C:\ComfyUI_windows_portable\ComfyUI\custom_nodes\SeargeSDXL\modules\stage_latent_inputs.py", line 127, in process (width, height) = get_mask_size(mask) File "C:\ComfyUI_windows_portable\ComfyUI\custom_nodes\SeargeSDXL\modules\utils.py", line 51, in get_mask_size (height, width) = mask.shape